Tropical Cyclone Genesis Ensemble Forecasts
Jiayi Peng*, Yuejian Zhu and Richard Wobus**IMSG at Environmental Modeling Center
Environmental Modeling Center /NCEP/NOAA, College Park, MD 20740
Acknowledgements: EMC Ensemble Team, EMC HWRF TeamGeophysical Fluid Dynamics Laboratory
National Hurricane Center, Joint Typhoon Warning CenterHurricane Forecast Improvement Project (HFIP)
Tropical Cyclone Research Forum(TCRF)/68th IHC, College Park, March 6, 2014
NOAA Seasonal Hurricane OutlookObserved numbers:
Named storms: 13Hurricanes: 2Major Hurricanes : 0ACE: 36% of median
Courtesy of Jon Gottschalck, CPC
Atlantic TC Genesis Verification(NHC,2013/2012)
2013: A slight under‐forecast (low) bias was present at the low to medium probabilities. Sample is small at the high probabilities.
A slight under‐forecast (low) bias was present in 2012.
Courtesy of James Franklin, NHC
East Pacific TC Genesis Verification(NHC,2013/2012)
2013: A slight low bias at all probabilities, but overall fairly well‐calibrated, andmuch improved from previous years.
2012: A low bias was present at the middle probabilities, and high bias at the high probabilities.
Courtesy of James Franklin, NHC
Ensemble ModelsModels Resolution Members Daily Frequency Forecast Length
NCEP ensemble GFS T254L42 -55km(02/14/2012)
20+1 00, 06, 12, 18 UTC
16 days (384hrs)
CMC ensemble GEM L72-66km(02/13/2013)
20+1 00, 12 UTC 16 days (384hrs)
ECMWF ensemble
IFS T639/319L62-30/60km
50+1 00, 12 UTC 15 days (360hrs)
FNMOC ensemble
NOGAPS T159L42-80km(09/14/2011)
20 00, 12 UTC 16 days (384hrs)
Our goal: Improve tropical‐cyclone genesis prediction by using NCEP, CMC , FNMOC and ECMWF global ensemble forecasts !
TC Genesis Target Area: 0‐40N, 110E‐350EForecast Time Window: 48‐hour genesis
NCEP GEFS TC Genesis Probability Forecast Within 48 hours
WP97 will form “DANAS” (WP23, 10/03 06Z)EP93 will form “NARDA”(EP14, 10/06 18Z)AL97 will form “KAREN” (AL12, 10/03 06Z)
AL97
EP93WP97
Algorithms for TC genesis probabilistic forecastHow to define global model TC genesis?The prediction vortices in Global Ensemble Forecast Systems are very weak. (25kts ?)
Step No.1: (for GEFS, ECMWF, CMC, FNMOC ensembles)We track every vortex by checking:1)850/700hPa/surface relative vorticity (max)2)850/700hPa geopotential height (min)3)Sea level pressure (min)4)850/700hPa/surface wind speed (min)5)SLP gradient (0.0015mb/km), Wind speed at 850hPa (≥ 1.5m/s)6)Closed SLP contour checked
Step No.2: (for GEFS and ECMWF ensemble)We filter those vortices based on the following criteria:1)Surface maximum wind speed ≥ 10kts2)850hPa maximum vorticity ≥ 10**(‐4) 1/s3)300‐500hPa temperature anomaly ≥ 0.5c
NHC Atlantic Basin Log: invest storm to TC genesis (June~November, 2013)No. INVEST Date TC No. INVEST Date TC
1 AL91 6/5 AL01 16 AL91 9/7,8 AL09
2 AL92 6/6‐7 17 AL92 9/8‐10
3 AL93 6/16,17 AL02 18 AL93 9/11,12 AL10
4 AL95 7/6,7,8 AL03 19 AL94 9/14‐16
5 AL96 7/11,12 20 AL95 9/17‐21
6 AL98 7/22,23,24 AL04 21 AL96 9/28,29 AL11
7 AL99 7/24,25 22 AL97 9/30,10/1‐3 AL12
8 AL91 7/28,29,8/1‐3 23 AL98 10/7‐13
9 AL92 8/15‐17 24 AL99 10/17
10 AL93 8/14,15 AL05 25 AL90 10/21 AL13
11 AL95 8/24,25 AL06 26 AL98 11/17‐18 AL14
12 AL96 8/30,31,9/1
13 AL97 8/31,9/1‐4 AL07
14 AL98 9/2‐3,5‐8
15 AL99 9/5,6 AL08
2013 Atlantic TC Genesis Forecast Reliability Diagram
0102030405060708090100
0 10 20 30 40 50 60 70 80 90 100Forecast (%)
Verifying
(%)
Red line: perfect reliability Green line: distribution of the forecast genesis probabilityBlue line: the relationship between the forecast and verifying genesis probability
OBSERVARTION
FORECAST YES NOYES a=22 b=48NO c=38 d=78Hit rate(POD)=a/(a+c) 37%
Miss rate=c/(a+c) 63%
False alarm rate=b/(b+d) 38%
Correct rejection rate=d/(b+d) 62%Critical success index=a/(a+b+c) 0.2
0102030405060708090100
0 10 20 30 40 50 60 70 80 90 100
GEFSN=186
Forecast (%)
OBSERVARTION
FORECAST YES NOYES a=17 b=33NO c=12 d=24
Hit rate(POD)=a/(a+c) 59%
Miss rate=c/(a+c) 41%
False alarm rate=b/(b+d) 58%
Correct rejection rate=d/(b+d) 42%Critical success index=a/(a+b+c) 0.27
ECMWFN=86
Contingent Table
2013 Atlantic TC Genesis Forecast Reliability Diagram
0102030405060708090100
0 10 20 30 40 50 60 70 80 90 100Forecast (%)
Verifying
(%)
OBSERVARTION
FORECAST YES NO
YES a=21 b=55
NO c=17 d=28
Hit rate(POD)=a/(a+c) 55%
Miss rate=c/(a+c) 45%
False alarm rate=b/(b+d) 66%
Correct rejection rate=d/(b+d) 34%
Critical success index=a/(a+b+c) 0.23
CMCN=121
0102030405060708090100
0 10 20 30 40 50 60 70 80 90 100Forecast (%)
FNMOCN=119
OBSERVARTION
FORECAST YES NO
YES a=21 b=44
NO c=16 d=37
Hit rate(POD)=a/(a+c) 57%
Miss rate=c/(a+c) 43%
False alarm rate=b/(b+d) 54%
Correct rejection rate=d/(b+d) 46%
Critical success index=a/(a+b+c) 0.26
NHC East Pacific Basin Log: invest storm to TC genesis (May~November, 2013)No. INVEST Date TC No. INVEST Date TC
0 EP90 5/13‐15 EP01 15 EP96 8/27,28 EP10
1 EP91 5/23‐30 16 EP97 8/28‐30
2 EP92 5/26‐28 EP02 17 EP98 8/29‐31 EP11
3 EP94 6/21‐23 EP03 18 EP99 9/3,4 EP12
4 EP95 6/21‐24 19 EP90 9/11‐13 EP13
5 EP96 6/28‐29 EP04 20 EP92 9/24‐28
6 EP97 7/2‐4 EP05 21 EP93 10/2‐6 EP14
7 EP98 7/21‐24 EP06 22 EP94 10/9‐13 EP15
8 EP99 7/29,30 EP07 23 EP95 10/12‐14 EP16
9 EP90 7/30,8/2,3 EP08 24 EP96 10/18‐20 EP17
10 EP91 8/2‐3 25 EP97 10/30,31,11/1 EP18
11 EP92 8/8‐15 26 EP98 11/14‐16
12 EP93 8/11
13 EP94 8/19‐22 EP09
14 EP95 8/25‐27
2013 East Pacific TC Genesis Forecast Reliability Diagram
0102030405060708090100
0 10 20 30 40 50 60 70 80 90 100Forecast (%)
Verifying
(%)
OBSERVARTION
FORECAST YES NO
YES a=42 b=15
NO c=72 d=109
Hit rate(POD)=a/(a+c) 37%
Miss rate=c/(a+c) 63%
False alarm rate=b/(b+d) 12%
Correct rejection rate=d/(b+d) 88%
Critical success index=a/(a+b+c) 0.33
GEFSN=238
0102030405060708090100
0 10 20 30 40 50 60 70 80 90 100Forecast (%)
OBSERVARTION
FORECAST YES NO
YES a=38 b=24
NO c=14 d=30
Hit rate(POD)=a/(a+c) 73%
Miss rate=c/(a+c) 27%
False alarm rate=b/(b+d) 44%
Correct rejection rate=d/(b+d) 54%
Critical success index=a/(a+b+c) 0.50
ECMWFN=106
2013 East Pacific TC Genesis Forecast Reliability Diagram
0102030405060708090100
0 10 20 30 40 50 60 70 80 90 100Forecast (%)
Verifying
(%)
OBSERVARTION
FORECAST YES NO
YES a=38 b=28
NO c=21 d=58
Hit rate(POD)=a/(a+c) 64%
Miss rate=c/(a+c) 36%
False alarm rate=b/(b+d) 33%
Correct rejection rate=d/(b+d) 67%
Critical success index=a/(a+b+c) 0.44
CMC N=145
0102030405060708090100
0 10 20 30 40 50 60 70 80 90 100Forecast (%)
FNMOCN=137
OBSERVARTION
FORECAST YES NO
YES a=21 b=9
NO c=35 d=72
Hit rate(POD)=a/(a+c) 38%
Miss rate=c/(a+c) 62%
False alarm rate=b/(b+d) 11%
Correct rejection rate=d/(b+d) 89%
Critical success index=a/(a+b+c) 0.32
West Pacific Basin Log: invest storm to TC genesis (June~December, 2013)No. INVEST Date TC No. INVEST Date TC
1 WP98 6/7,8 WP03 21 WP97 9/11,12 WP16
2 WP99 6/14,15 22 WP99 9/16 WP17
3 WP91 6/16,17 WP04 23 WP90 9/16,17,18 WP18
4 WP93 6/17 24 WP98 9/19,20,21 WP19
5 WP94 6/19,20 WP05 25 WP91 9/20
6 WP99 6/27,28 WP06 26 WP93 9/25,26 WP20
7 WP92 7/7 WP07 27 WP96 9/29 WP21
8 WP93 7/14,15 WP08 28 WP95 9/27‐30 WP22
9 WP94 7/18,19 29 WP97 10/2,3 WP23
10 WP91 7/30,31 WP09 30 WP90 10/7‐9
11 WP93 8/1 31 WP91 10/8 WP24
12 WP94 8/4,5 WP10 32 WP92 10/8,9,10 WP25
13 WP96 8/8 WP11 33 WP93 10/15,16 WP26
14 WP98 8/16,17 WP12 34 WP94 10/18,19 WP27
15 WP99 8/16,17 WP13 35 WP95 10/19,20 WP28
16 WP91 8/24,25 WP14 36 WP96 10/28,29 WP29
17 WP92 8/28 37 WP98 11/2,3 WP30
18 WP93 8/28‐30 38 WP99 11/3 WP31
19 WP95 9/9,10 39 WP90 11/10‐14 WP32
20 WP96 9/1 WP15 40 WP92 11/21‐23
41 WP94 12/3 WP33
2013 West Pacific TC Genesis Forecast Reliability Diagram
0102030405060708090100
0 10 20 30 40 50 60 70 80 90 100Forecast (%)
Verifying
(%)
OBSERVARTION
FORECAST YES NO
YES a=84 b=20
NO c=29 d=35
Hit rate(POD)=a/(a+c) 74%
Miss rate=c/(a+c) 26%
False alarm rate=b/(b+d) 36%
Correct rejection rate=d/(b+d) 64%
Critical success index=a/(a+b+c) 0.63
GEFSN=168
0102030405060708090100
0 10 20 30 40 50 60 70 80 90 100Forecast (%)
ECMWFN=83
OBSERVARTION
FORECAST YES NO
YES a=44 b=6
NO c=11 d=22
Hit rate(POD)=a/(a+c) 80%
Miss rate=c/(a+c) 20%
False alarm rate=b/(b+d) 21%
Correct rejection rate=d/(b+d) 79%
Critical success index=a/(a+b+c) 0.72
2013 West Pacific TC Genesis Forecast Reliability Diagram
0102030405060708090100
0 10 20 30 40 50 60 70 80 90 100Forecast (%)
Verifying
(%)
OBSERVARTION
FORECAST YES NO
YES a=56 b=23
NO c=13 d=19
Hit rate(POD)=a/(a+c) 81%
Miss rate=c/(a+c) 20%
False alarm rate=b/(b+d) 55%
Correct rejection rate=d/(b+d) 45%
Critical success index=a/(a+b+c) 0.61
CMC N=111
0102030405060708090100
0 10 20 30 40 50 60 70 80 90 100Forecast (%)
OBSERVARTION
FORECAST YES NO
YES a=41 b=16
NO c=28 d=24
Hit rate(POD)=a/(a+c) 59%
Miss rate=c/(a+c) 41%
False alarm rate=b/(b+d) 40%
Correct rejection rate=d/(b+d) 60%
Critical success index=a/(a+b+c) 0.48
FNMOC N=109
Next Ensemble (GEFS) Implementation• Model
– Current: GFS Euler model– Plan: GFS Semi‐Lagrangian model
• Horizontal resolution– Current: T254 (55km for 0‐192 hours), T190 (73km for 192‐384 hours) – Plan: Variable resolution
– T574 (T382 physics ‐ 34km for 0‐168 hours)– T382 (T254 physics – 55km for 168‐384 hours
• Vertical resolution– Current: L42 hybrid levels– Plan: L64 hybrid levels (will match with GFS and DA)
• Ensemble memberships:– The same as current operational – 21 members for each cycle (additional discussion)
• Computation cost:– Current: 84 nodes (+ post process) for 55 minutes– Plan: 252 nodes (+ post process) for peak time, within one hour for integration
• Initial perturbations:– Current: BV‐ETR 6‐hr cycling– Plan: EnKF f06 + 3DETR
• Stochastic perturbations:– Current: STTP– Plan: tuned STTP for new model, initial perturbations and higher resolution
• Output:– Current: every 6‐hr for 1*1 degree pgrb files– Plan: every 3‐hr for 0.5*0.5 degree pgrb files
• Challenge:‐ T574L64 configuration will need nearly 300 nodes in peak time• Expectations:‐ Improving probabilistic forecast skills overall‐ Improving Tropical Storm track forecast errors, and intensity forecast
Summary
1) Four ensembles over‐predicted Atlantic TC genesis in 2013.2) Four ensembles had reliable forecasts for East Pacific TC genesis, and
higher hit rate in West Pacific TC genesis in 2012.3) The new GEFS (GFS T574L64 Ensemble) implementation will provide more
reliable TC genesis guidance for the coming hurricane season.
Please visit:http://www.emc.ncep.noaa.gov/gmb/jpeng/TC_ens_V1.html
for 2014 real time TC track and genesis probabilistic forecasts.
2013 GFS TC Genesis Forecast Contingent Table
OBSERVARTION
FORECAST YES NO
YES a=19 b=41
NO c=16 d=34
Hit rate(POD)=a/(a+c) 54%
Miss rate=c/(a+c) 46%
False alarm rate=b/(b+d) 55%
Correct rejection rate=d/(b+d) 45%
Critical success index=a/(a+b+c) 0.25
Atlantic, N=110OBSERVARTION
FORECAST YES NO
YES a=45 b=25
NO c=16 d=20
Hit rate(POD)=a/(a+c) 74%
Miss rate=c/(a+c) 26%
False alarm rate=b/(b+d) 56%
Correct rejection rate=d/(b+d) 44%
Critical success index=a/(a+b+c) 0.52
East Pacific, N=106
OBSERVARTION
FORECAST YES NO
YES a=86 b=23
NO c=7 d=17
Hit rate(POD)=a/(a+c) 92%
Miss rate=c/(a+c) 8%
False alarm rate=b/(b+d) 58%
Correct rejection rate=d/(b+d) 42%
Critical success index=a/(a+b+c) 0.74
West Pacific, N=133
GFS T574L64 -27kmMax wind speed>=25kts
TC Genesis Tracker Code
Based on Tim Marchok’s latest version for genesis track, we make the following changes: (1)Calculation for 300‐500hPa temperature anomaly;(2)Calculation for 200‐850hPa zonal wind shear;(3)Calculation for 500hPa mean relative humidity;(4)The AL90‐99, EP90‐99 and WP90‐99 are those invest storms
named by NHC and JTWC. In case of NHC and JTWC miss potential TC genesis, we create some interested storms based on GFS‐T574 operational, Canadian, NOGAPS and ECMWF Deterministic Forecasts, named as HC01, HC02, etc.
Algorithms for TC genesis probabilistic forecast
TC Genesis: maximum sustainable winds ≥25kts (12.9m/s)Criteria (Gray,1968):1)A positive relative vorticity in lower troposphere2)Far away 5N/S from equator3)Warm SST (≥79F or 26.1C for Atlantic)4)Small vertical shear5)Conditional instability in the lower to mid‐troposphere6)Large value of relative humidity in the middle troposphere
TC Genesis Target Area: 0‐40N, 110E‐350EForecast Time Window: 48‐hour genesis
WP92 EP95
HC03
HC05
NCEP GEFS TC Genesis Probability Forecast Within 48 hours
WP92 will form “Talim” (WP06, 06/18 00Z)HC03 will form “Chris”(AL03, 06/18 18Z)